Table4_Machine learning-based identification of a consensus immune-derived gene signature to improve head and neck squamous cell carcinoma therapy and outcome.XLS

BackgroundHead and neck squamous cell carcinoma (HNSCC), an extremely aggressive tumor, is often associated with poor outcomes. The standard anatomy-based tumor–node–metastasis staging system does not satisfy the requirements for screening treatment-sensitive patients. Thus, an ideal biomarker leading to precise screening and treatment of HNSCC is urgently needed.
MethodsTen machine learning algorithms—Lasso, Ridge, stepwise Cox, CoxBoost, elastic network (Enet), partial least squares regression for Cox (plsRcox), random survival forest (RSF), generalized boosted regression modelling (GBM), supervised principal components (SuperPC), and survival support vector machine (survival-SVM)—as well as 85 algorithm combinations were applied to construct and identify a consensus immune-derived gene signature (CIDGS).
ResultsBased on the expression profiles of three cohorts comprising 719 patients with HNSCC, we identified 236 consensus prognostic genes, which were then filtered into a CIDGS, using the 10 machine learning algorithms and 85 algorithm combinations. The results of a study involving a training cohort, two testing cohorts, and a meta-cohort consistently demonstrated that CIDGS was capable of accurately predicting prognoses for HNSCC. Incorporation of several core clinical features and 51 previously reported signatures, enhanced the predictive capacity of the CIDGS to a level which was markedly superior to that of other signatures. Notably, patients with low CIDGS displayed fewer genomic alterations and higher immune cell infiltrate levels, as well as increased sensitivity to immunotherapy and other therapeutic agents, in addition to receiving better prognoses. The survival times of HNSCC patients with high CIDGS, in particular, were shorter. Moreover, CIDGS enabled accurate stratification of the response to immunotherapy and prognoses for bladder cancer. Niclosamide and ruxolitinib showed potential as therapeutic agents in HNSCC patients with high CIDGS.
ConclusionCIDGS may be used for stratifying risks as well as for predicting the outcome of patients with HNSCC in a clinical setting.

背景：头颈部鳞状细胞癌（Head and neck squamous cell carcinoma, HNSCC）是一种极具侵袭性的肿瘤，往往预后不佳。当前基于解剖结构的肿瘤-淋巴结-转移（tumor-node-metastasis, TNM）分期系统无法满足筛选治疗敏感型患者的需求，因此亟需一种可实现头颈部鳞状细胞癌精准筛查与治疗的理想生物标志物。
方法：本研究采用10种机器学习算法——套索回归（Lasso）、岭回归（Ridge）、逐步Cox回归（stepwise Cox）、CoxBoost、弹性网络（elastic network, Enet）、Cox偏最小二乘回归（partial least squares regression for Cox, plsRcox）、随机生存森林（random survival forest, RSF）、广义提升回归模型（generalized boosted regression modelling, GBM）、监督主成分分析（supervised principal components, SuperPC）以及生存支持向量机（survival support vector machine, survival-SVM）——连同85种算法组合，用于构建并鉴定共识免疫源性基因特征（consensus immune-derived gene signature, CIDGS）。
结果：基于包含719名头颈部鳞状细胞癌患者的三个队列的表达谱数据，我们通过上述10种机器学习算法与85种算法组合，筛选得到236个共识预后基因，并进一步提炼为共识免疫源性基因特征（CIDGS）。训练队列、两个测试队列以及荟萃队列的分析结果均一致表明，CIDGS能够精准预测头颈部鳞状细胞癌患者的预后。纳入若干核心临床特征与51种已报道的基因特征后，CIDGS的预测能力得到显著提升，优于其他各类基因特征。值得注意的是，低CIDGS评分患者的基因组改变更少、免疫细胞浸润水平更高，对免疫治疗及其他治疗药物的敏感性更强，同时预后更佳；而高CIDGS评分的头颈部鳞状细胞癌患者生存时间往往更短。此外，CIDGS还能够准确区分膀胱癌患者的免疫治疗响应与预后情况。研究发现，氯硝柳胺（Niclosamide）与芦可替尼（ruxolitinib）有望作为高CIDGS评分头颈部鳞状细胞癌患者的治疗药物。
结论：在临床实践中，共识免疫源性基因特征（CIDGS）可用于对头颈部鳞状细胞癌患者进行风险分层与预后预测。